The Facets and Consequences of Uncertainty in Human–AI Interaction

Abstract

Abstract Artificial intelligence (AI) is increasingly used to assist or substitute humans in organizational processes. The transfer of decision-making power and task performance from humans to artificial agents increases uncertainty—defined as a state of not knowing stemming from incomplete and/or ambiguous information—for workers in several ways. Within the broader context of environmental and societal uncertainty, the authors focus specifically on uncertainty at the task level and develop a framework about uncertainty in human–AI interaction, showing that people experience three main facets of uncertainty in their interactions with AI: (a) usage uncertainty—not knowing what specific tasks AI can perform; (b) process uncertainty—not knowing how AI performs tasks and achieves its results; and (c) outcome uncertainty—not knowing what the outcome of AI’s processes will be and how to evaluate them. The authors discuss how each facet of uncertainty can lead to algorithm aversion and appreciation, as well as what can be done to change attitudes and behaviors toward AI from an uncertainty lens. The authors conclude by providing an outlook on how the “AI wave” is changing the nature of work and engendering fears of replacement due to the automation risk, and they ultimately suggest the development of joint cognitive systems between humans and technology to fruitfully regulate uncertainty in human–AI interaction.